Evolving Autonomous Charging Behavior in a Robot Swarm

Abstract

Long-term autonomous operation of a robotic swarm in harsh or inaccessible locations will require the ability of the swarm to manage operational details such as battery charging without human intervention. Small robots may not be able to manage their own power internally, via solar cells or fuel cells, but may be dependent on external means to recharge batteries. In this paper we will demonstrate the ability of a simple robotic swarm to use a genetic algorithm (GA) to evolve optimized, behavioral parameters including the ability to determine battery charging behavior. We introduce two new battery parameters to the robots’ central place foraging algorithm (CPFA) and allow the GA to optimize the behaviors such that, on average, no robots are left “dead” due to inadequate battery charge. We also demonstrate that the GA can evolve a successful resource collection strategy while minimizing the number of robot casualties. The additional parameters and behaviors do not alter the original algorithm’s simplicity, ability to run in real-time, and requirement of using only local knowledge.